Pharmacogenomic and pharmacoproteomic approaches are creating major opportunities for progress in rational drug discovery and individualization of therapy for cancer patients. That much is clear. But taking advantage of the potential has proved more difficult than expected. One reason is that no single type of molecular information captures all of the relevant pharmacological and toxicological phenomena. Data on mRNA expression, protein expression, post-translational modifications, DNA copy number, SNP's, chromosomal aberrations, transcription factors, RNA-mediated regulation, CpG methylation, and molecular interactions of many types can all contribute. We need a research strategy that generates and integrates all of those types of information and more. The Genomics & Bioinformatics Group (GBG) is developing such a strategy, based on an integrative systems orientation and on synergy between traditional hypothesis-driven and 'omic' research. This project has been considered by at least one leader in the Human Cancer Genome Project as presaging that enterprise and providing useful background expertise for it. We have had to confront, in microcosm, many of the same issues - related to harvesting and quality-control of cancer samples, optimization of purification methods, optimization of experimental platforms, primary data analysis, integration of different data types, statistical and machine-learning multivariate analysis, biological interpretation, and management of relationships as the hub of collaborative interactions with numerous high-profile investigators and institutions.The GBG's studies are part experimental, part computational. Although we're ultimately interested in clinical tumors, many of our experiments are done in vitro with isogenic cell sets, with selected resistant variants, or with the 60 human cancer cell types (the NCI-60) used by the NCI Developmental Therapeutics Program to screen >100,000 chemical compounds for anticancer activity since 1990. Largely through studies by the GBG and collaborators, the NCI-60 are by far the most comprehensively profiled panel of mammalian cells anywhere. Cells in culture don't reflect the complexity of the in vivo environment, of course, but they circumvent the logistical issues, lack of homogeneity, lack of reproducibility, and artifacts of clinical materials. When hypotheses based on cell line data relate to the clinic, we often test them using tissue arrays. Five of our findings with clinical implications are (i) identification of biomarkers to distinguish colon from ovarian tumors of unknown origin, (ii) identification of asparagine synthetase as a predictor of ovarian cancer cell response to L-asparaginase; (iii) formulation of the "Permissive-Apoptosis Resistance" (PAR) two-step model for development of acquired drug resistance; (iv) discovery of 'MDR1-inverse' compounds more active in cancer cells that express large amounts of MDR1; (v) critical evidence in the 1990's that led to clinical development of oxaliplatin, now a 'standard-of-care' agent for therapy of both primary and recurrent colorectal cancer. We're translating our ideas and strategies into the clinical realm through collaborations with Curtis Harris, Elise Kohn, Yves Pommier, David Nelson, Tito Fojo, and Dan Von Hoff.Our large-scale molecular databases profiling the NCI-60:-Protein level: 2-D gel database of 1014 spots (with N.L. Anderson); 648-spot "reverse-phase" protein lysate arrays, run for >240 proteins to date (with Liotta/Petricoin).-RNA level: 9706-spot cDNA arrays (with Brown/Botstein); 6800-gene Affymetrix arrays (with Lander/Golub); 60,000-gene U95 and 30,000-gene U133 Affymetrix arrays (with Scherf/ Dolginow, Gene Logic); real-time RT-PCR for the 48 ABC transporters (with M. Gottesman); transporter oligo array (with W. Sadee); ABC-ToxChip (with Gottesman/Annereau).-DNA level: Affymetrix chips of 1600 SNPs (with K. Buetow), SKY (with I. Kirsch); BAC array CGH (with J. Gray)